The relationship between forest cover and streamflow of watersheds is complex and still controversial in the scientific literature. To investigate suchrelationship we propose an alternative method which requires the following information for each watershed: percentage of forest cover, annual rainfall, average specific streamflow (qave), and minimum mean specific streamflow in seven consecutive days (q7). As a case study, we analyzed a dataset composed by 25 watersheds located in the Espírito Santo State (ESS), Brazil. We conducted simple and multiple linear regression analyses as well as partial correlation analysis between the above parameters. To reduce the effect of heterogeneity of environmental factors, watersheds with similar characteristics in term of rainfall, drainage area, and both rainfall and drainage area were grouped by cluster analysis, and the above regression and correlation analysis was repeated on each group. Our results using the whole dataset showed that forest cover has a negative relationship with watershed streamflow. The analysis of homogeneous groups of watersheds showed that the average minimum streamflow during seven days (q7) was more sensitive to the presence of forest cover, showing a negative relationship, especially in watersheds with low annual rainfall, while in areas with high precipitation, the annual rainfall showed a strong influence on the hydrological responses of watersheds, masking the effect of forest cover. The proposed method may be easily extended to other areas, and allowsthe inclusion of other relevant environmental variables according to specific cases.
Land-use intensity is a relevant factor of land cover change, which is leading many developing countries to experience the depletion of natural resources (
The ecological and hydrological imbalances caused by land-use changes in watersheds are widely recognized in the scientific community (
Considering the above-mentioned uncertainties, studies that investigate the effects of forests on hydrological streamflows are necessary. So far, these studies have been generally carried out using paired experimental watersheds or long-term time-trends. Paired studies focus on two watersheds that are close and similar in terms of environmental, physical, and climatic aspects. Most of the experiments (
A significant portion of the current knowledge on the hydrological response of streamflows to forest management comes from studies relyingon the above consolidated methods, which have been conducted worldwide (
Studies that correlate forest cover with streamflow have been conducted in monitored watersheds for several years, where hydrological data are obtained before and after the forest management (
In this study, we propose an alternative methodology based on simple statistics where neither forest suppression nor the manipulation of large volumes of historical data is required. Both rainfall and watershed size are included as influencing factors in the hydrological response. Several watersheds are used to improve regional representativeness and allow the simultaneous analysis of large areas in different regions. The low cost and easy applicability of this method can mainly benefit regions with a lack of resources to implement paired watershed experiments. Additionally, this approach is also beneficial to regions where the development of long-term continuous monitoring of hydrological and forestry variables is challenging.
To test the proposed methodology, we conducted our study in Southeastern Brazil, which is historically affected by strong regional differences in the volume of available water and has experienced worrying drought periods and water scarcity in recent years (
The unprecedented drought in Brazil may be a direct consequence of inflow depletion from the Amazon watershed, which normally brings rainfall to Midwest and Southeastern Brazil (
In this study, we focused on the relationship between streamflow and forest cover using environmental and empirical hydrological data from governmental agencies. The main goal was to test a new method to efficiently determine the correlation between forest cover and streamflow in watersheds and its application to other regions. We finally propose the adopted methodologyas a viable alternative to long years of watershed monitoring data or suppression management.
We used data recorded from several watersheds in one typical hydrological year. The short period of analysis is compensated by studying several watersheds at the same time under different environmental and land use conditions, which allows a better regional representation. The dataset included the following types of information: the percentage of forest cover, annual rainfall, average specific streamflow, and minimum average specific streamflow in seven consecutive days. Simple and multiple regression analysis as well as partial correlation were applied to disentangle the effect of forest cover and rainfall on watershed streamflow. To reduce the effect of heterogeneity of environmental factors, watersheds with similar characteristics in term of rainfall, drainage area, and both rainfall and drainage area were grouped by cluster analysis, and the above regression analysis was repeated on each group. A flowchart summarizing the proposed methodology is reported in
We selected 25 watersheds located in the Espírito Santo State (ESS), Southeastern Brazil (
We focused on records from the streamflow and rainfall data platforms of the ESS, which were collected in the hydrological year 2007/2008 (beginning in October 2007 and ending in September 2008). This dataset was chosen since it was the most recent and complete survey of land use of the ESS (see below) and provides information from previous and subsequent years. The average annual rainfall for the selected period (1074 mm) was within the normal long-term range for ESS (1186 ± 210 mm year-1), indicating that the selected year is representative of local historical conditions. Additionally, we compared the geographical distribution and patterns of rainfall for the period 2007/2008 with the historical records to confirm its hydrological representativeness (
We delimited the watersheds by referring to the upstream drainage area of each streamflow station using a Hydrologically-Consistent Digital Elevation Model (HCDEM). The HCDEM was created from the Digital Elevation Model (DEM) of SRTM (Space Shuttle Radar Topographic Mission), which was obtained from the United States Geological Survey (USGS - https://earthexplorer.usgs.gov/) with a 30-meter resolution. Firstly, we created a mosaic of images and filled DEM sinks. To improve terrain representation, specifically regarding hydrological consistency, we reconditioned the DEM and created the HCDEM using the AGREE algorithm (
The watershed delimitation was obtained by successive applications of the following tools of the HCDEM in the Hydrology toolbox in ArcMap®: fill, flow direction, flow accumulation, stream definition, snap pour point, and watershed. We used the river station locations at the mouths of the main rivers as references for allocating the pour points in the snap pour point tool. Finally, the automatic delimitation of the watersheds under study was performed using the watershed tool.
The streamflow data for the hydrological year 2007/2008 was obtained from streamflow stations of the ESS, which are freely available at the National Information System on Water Resources (
The average annual streamflow rate (Qave) and the average minimum streamflow rate for seven days (Q7) for the hydrological year 2007/2008 were obtained for each station. We also estimated the respective specific streamflows (qave and q7), which are calculated as the ratio between the streamflow (Q) and the drainage area (km²) of each watershed.
Daily rainfall data was obtained from the daily gridded meteorological variables in Brazil (https://utexas.app.box.com/v/Xavier-etal-IJOC-DATA -
Forest cover data for the years 2007 and 2008 were obtained from the aerophotogrammetric land use survey carried out by the
The relationships of the percentage of forest cover and the rainfall with the average specific annual streamflow (qave) and the minimum specific streamflow with seven days duration (q7) were estimated using simple and multiple linear regression. Also, a partial correlation between forest cover and streamflow was applied using the rainfall as a fixed effect. The analyses were performed using together all the 25 watersheds, and the significance of the results was assessed by the F-test with α = 0.10.
As watersheds naturally presented different sizes, the contrasting rainfall regimes may influence their hydrological behavior. Thus, the linear regression and the partial correlation were applied separately after grouping the watersheds by similarity. Three different approaches of watershed grouping were adopted according to: (a) homogeneous regions of rainfall; (b) drainage area; and (c) both rainfall and drainage area. We used the hierarchical cluster analysis along with Ward’s method (
The results obtained from the data of all 25 watersheds are presented in
The relationship between streamflow and forest cover using the whole dataset of 25 watershed was not significant (
Cluster analysis based on rainfall similarity allowed to detect four groups of watersheds, considering the assumed cut point (
Cluster analysis of watersheds based on their drainage area resulted in four groups (
Finally, the cluster analysis carried out according to both drainage areas and rainfall allowed to detect four groups of watersheds. The group formed by the Coutinho and Usina Paineiras watersheds was discarded due to the insufficient number of watersheds. The remaining three groups (AP1, AP2, and AP3) were as follows (
A significant association between minimum streamflow and rainfall (p-value = 0.036) was found for group P2, which included the watersheds with intermediate precipitation, with a negative tendency (rp = -0.84). This means that an increase in rainfall imply a reduction in the streamflow, thus contrasting the physical processes of inflow and outflow of water in the watershed. However, the environmental heterogeneity of the region, which is composed of coastal to mountainous areas, may have a great influence on local evapotranspiration, infiltration, and water storage rates due to the presence of other environmental factors not analyzed in this work, such as geological or pedological aspects. Moreover, in some cases, environmental factors may not directly explain changes in the streamflow regime but may be an effect of anthropogenic influences (
Regarding the influence of forest cover in the first group by rainfall (P1, P2, and P3), we found a significantrelationship with average and minimum streamflows (
According to
A widely discussed hypothesis in the scientific community about the relationship between forests and streamflows is the “infiltration-evapotranspiration trade-off”, postulated by Bruijnzeel (
In this study, we observed higher streamflows in less forested areas and
The reduction in water infiltration into soil restricts the replenishment of groundwater (
Increases in streamflows due to decreased forest cover can be associated with water losses through forest evapotranspiration (
The negative association between forest cover and streamflows is likely a consequence of water consumption by vegetation. This event is significant during drought or low rainfall periods, and affects the water availability of watersheds under these circumstances. Regarding the negative relationship between forest cover and water availability, similar results were obtained in other studies under different methodologies and different vegetation types worldwide (
Regarding the difference between these results and the analysis of the 25 watersheds, besides the rainfall rate, we believe that this pattern is an effect of watersheds size, which may influence the hydrological response. Afterwards, the effect of the spatial scale on hydrological responses caused by changes in forest cover is still poorly understood and inconclusive. However, it is known that large and small watersheds can present different hydrological responses to the same factor analyzed (
Regression analysis carried out for the groups A1, A2, and A3 showed that only rainfall significantly affected the streamflow, while the percentage of forest cover did not have a significant influence on streamflow (
Our results showed that the percentage of forest cover had a significant influence on streamflow when the watershed groups based on drainage area and rainfall (AP1, AP2, and AP3) are considered. Besides, we found a significant influence of forest cover on the average streamflow (qave) for the AP2 group (varied sizes and lower rainfall) using a simple linear regression (R2 = 0.45, p-value = 0.049 -
The annual runoff is more sensitive to forest cover at different spatial scales in watersheds with limited rainfall, causing more significant hydrological responses (
The effect of forests on flow patterns in the dry season is one of the most contradictory aspects in forest hydrology, with conflicting evidence for different combinations of forests, rainfall, and soil conditions. During the drought season, the effect of the balance between infiltration capacity and evapotranspiration becomes even more prominent (
We observed that all the significant partial correlation coefficients between streamflow and forest cover were negative, which means that smaller streamflows occurred in watersheds with a higher percentage of forest cover and
Based on the dynamic of the balance between infiltration and evapotranspiration, we emphasize the importance of prioritizing good soil management practices among the land uses that already exist in watersheds. For example, reforestation of watersheds should be well planned considering the hydrological and soil conditions at each site. Indeed, the replacement of other land uses with forested areas may not always reflect substantial gains in streamflows (
Although forestsdo not always increasethe annual water yield in watersheds, it can affect other important hydrological mechanisms. For example, forest loss can weaken the regulating mechanisms of tropical watersheds, altering river flow regimes and possibly leading to extreme events, such as floods and/or water shortage (
Large-scale deforestation in tropical rainforests can influence the tropospheric moisture flows and atmospheric processes that regulate the transition between dry and rainy seasons. According to observations of reciprocal feedbacks on climate-forest for the Amazon, deforestation can delay the onset of rainfall at regional scale, as discussed by
Water security for human society is a real need and a worldwide concern. On the other hand, population growth and the development of productive activities are closely related to land-use changes and the conversion of forest areas.
In this study, we analyzed the relationships between streamflow, rainfall and forest cover over the period 2007/2008 in the Espirito Santo State, southeastern Brazil, by grouping watersheds with similar physical and/or environmental characteristics. This approach can help local managers to better understand how forest management can reduce or increase hydrological availability.
The average minimum streamflow during seven days (q7) was more sensitive to the presence of forest cover, showing a negative relationship, especially in watersheds with low annual rainfall. Regarding the areas with high rainfall levels, the annual rainfall showed a strong influence on the hydrological responses of watersheds, regardless of the percentage of forest cover.
The methodology applied in this study is a viable and easy-to-apply alternative to (but not a substitute of) consolidated methods in hydrological science, such as experimental watersheds and long-term time-trends. Our approach can be usefully replicated in other regions, as long as the hydrological variables for the analyzed period fall within their normal historical range, since it does not require the experimental suppression of forest cover and allows the simultaneous study of several watersheds, which can be very useful for water management. Moreover, other environmental variables related to the hydrological dynamics of watersheds may be included in the analysis, aimed to throw light in greater detail on this complex association.
We thank the National Council for Scientific and Technological Development (CNPq) for the financial support of this study.
Flowchart of the proposed method. Firstly, regression and partial correlation analysis were applied on the whole data set of 25 watersheds. Cluster analysis was then performed to identify groups of watersheds with similar environmental characteristics. Finally, regression and correlation analyses were repeated separately for each group of watersheds.
Geographical pattern of the rainfall in the Espírito Santo State (ESS). (Left panel): rainfall distribution in the hydrological year 2007/2008; (right panel): historical mean rainfall distribution.
Dendrograms obtained by cluster analysis of watersheds with similar environmental characteristics, according to: (a) rainfall; (b) drainage area; and (c) both drainage area and rainfall.
Data from the watersheds under study in Espírito Santo State, Brazil. (qave): average specific annual streamflow (L s-1 km-2); (q7): minimum specific streamflow with seven days duration (L s-1 km-2).
ID | Watershed name | Area(km²) | Streamflow | Rainfall(mm) | Forestcover (%) | |
---|---|---|---|---|---|---|
qave | q7 | |||||
1 | Pedro Canário | 1665.9 | 3.1 | 0.3 | 690.8 | 4.7 |
2 | São Jorge da Barra Seca | 451.7 | 4.6 | 1.0 | 820.3 | 15.7 |
3 | Laranja da Terra | 1331.7 | 10.1 | 4.1 | 1023.2 | 23.4 |
4 | Baixo Guandú | 2143.2 | 4.7 | 1.1 | 908.0 | 25.8 |
5 | Córrego da Piaba | 879.4 | 3.7 | 0.2 | 818.8 | 28.5 |
6 | Ponte do Pancas | 919.3 | 3.6 | 0.2 | 814.3 | 21.7 |
7 | São Gabriel da Palha | 1029.4 | 8.2 | 1.2 | 848.8 | 15.7 |
8 | Valsugana Velha - Montante | 90.3 | 13.8 | 0.5 | 1010.1 | 58.0 |
9 | Santa Leopoldina | 1011.6 | 7.2 | 3.0 | 1042.2 | 48.5 |
10 | Córrego do Galo | 979.0 | 10.9 | 5.0 | 1218.8 | 40.2 |
11 | Fazenda Jucuruaba | 1688.6 | 11.5 | 5.3 | 1241.8 | 45.7 |
12 | Matilde | 207.3 | 23.1 | 9.4 | 1480.9 | 58.2 |
13 | Usina Fortaleza | 223.0 | 12.1 | 3.1 | 1282.6 | 24.0 |
14 | Iúna | 433.5 | 14.4 | 5.8 | 1257.8 | 11.0 |
15 | Terra Corrida - Montante | 594.0 | 13.4 | 5.3 | 1281.2 | 12.0 |
16 | Itaici | 1047.4 | 13.1 | 4.1 | 1303.4 | 17.1 |
17 | Ibitirama | 341.6 | 28.1 | 5.2 | 1354.2 | 30.1 |
18 | Rive | 2221.0 | 14.7 | 4.9 | 1355.0 | 18.5 |
19 | Pacotuba | 2759.6 | 13.6 | 3.7 | 1358.5 | 18.2 |
20 | Fazenda Lajinha | 436.2 | 12.4 | 2.6 | 1305.7 | 33.9 |
21 | Castelo | 976.1 | 14.0 | 2.3 | 1364.7 | 30.3 |
22 | Usina São Miguel | 1457.5 | 15.1 | 3.3 | 1403.1 | 31.9 |
23 | Coutinho | 4604.4 | 14.9 | 3.7 | 1377.8 | 23.1 |
24 | Usina Paineiras | 5169.3 | 14.2 | 3.9 | 1377.8 | 22.6 |
25 | Guaçuí | 411.9 | 25.0 | 8.2 | 1452.5 | 22.1 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Simple linear regression between streamflow, forest cover, and rainfall for all 25 watersheds. (qave): average specific annual streamflow (L s-1km-2); (q7): minimum specific streamflow with seven days duration (L s-1 km-2); (R2): coefficient of determination.
Variable | Stats | Rainfall | Forest cover |
---|---|---|---|
qave | R² | 0.68 | 0.06 |
p-value | <0.001 | 0.228 | |
q7 | R² | 0.61 | 0.04 |
p-value | <0.001 | 0.346 |
Multiple linear regression and partial correlation between streamflow, forest cover, and rainfall for all 25 watersheds; (qave): average specific annual streamflow (L s-1 km-2); (q7): minimum specific streamflow with seven days duration (L s-1 km-2); (rp): coefficient of partial correlation; (R2): coefficient of determination.
Variable | Stats | Rainfall | Forest cover |
---|---|---|---|
qave | R² | 0.69 | |
p-value | <0.001 | ||
rp | 0.82 | 0.14 | |
p-value | <0.001 | 0.516 | |
q7 | R² | 0.61 | |
p-value | <0.001 | ||
rp | 0.77 | 0.06 | |
p-value | <0.001 | 0.8 |
Simple linear regression between streamflow, forest cover, and rainfall for the watershed groups obtained by cluster analysis. (qave): average specific annual streamflow (L s-1 km-2); (q7): minimum specific streamflow with seven days duration (L s-1 km-2); (R2): coefficient of determination.
ClusterAnalysis | Watershedgroups | Variables | Streamflow qave | Streamflow q7 | ||
---|---|---|---|---|---|---|
R² | p-value | R² | p-value | |||
By rainfall | P1 | Rainfall | 0.21 | 0.218 | 0.68 | 0.006 |
Forest | 0.16 | 0.285 | 0.30 | 0.126 | ||
P2 | Rainfall | 0.22 | 0.285 | 0.41 | 0.120 | |
Forest | 0.82 | 0.005 | 0.03 | 0.710 | ||
P3 | Rainfall | 0.23 | 0.333 | 0.37 | 0.202 | |
Forest | 0.00 | 0.981 | 0.00 | 0.971 | ||
By drainage area | A1 | Rainfall | 0.95 | <0.001 | 0.62 | 0.037 |
Forest | 0.17 | 0.365 | 0.36 | 0.155 | ||
A2 | Rainfall | 0.88 | 0.002 | 0.64 | 0.032 | |
Forest | 0.00 | 0.986 | 0.15 | 0.390 | ||
A3 | Rainfall | 0.60 | 0.014 | 0.71 | 0.004 | |
Forest | 0.09 | 0.437 | 0.00 | 0.962 | ||
By rainfall and drainage area | AP1 | Rainfall | 0.94 | <0.001 | 0.52 | 0.069 |
Forest | 0.39 | 0.136 | 0.09 | 0.525 | ||
AP2 | Rainfall | 0.57 | 0.018 | 0.50 | 0.033 | |
Forest | 0.45 | 0.049 | 0.03 | 0.637 | ||
AP3 | Rainfall | 0.57 | 0.048 | 0.65 | 0.028 | |
Forest | 0.16 | 0.376 | 0.14 | 0.400 |
Multiple linear regression and partial correlation between streamflow, forest cover, and rainfall for the watershed groups obtained by cluster analysis. (qave): average specific annual streamflow (L s-1 km-2); (q7): minimum specific streamflow with seven days duration (L s-1 km-2); (R2): coefficient of determination; (rp): coefficient of partial correlation.
ClusterAnalysis | Watershedgroups | Variables | Streamflow qave | Streamflow q7 | ||||||
---|---|---|---|---|---|---|---|---|---|---|
R² | p-value | rp | p-value | R² | p-value | rp | p-value | |||
By rainfall | P1 | Rainfall | 0.22 | 0.469 | 0.27 | 0.512 | 0.68 | 0.032 | 0.74 | 0.036 |
Forest | 0.14 | 0.740 | -0.03 | 0.949 | ||||||
P2 | Rainfall | 0.82 | 0.032 | 0.10 | 0.858 | 0.72 | 5.047 | -0.84 | 0.036 | |
Forest | -0.88 | 0.022 | -0.72 | 0.107 | ||||||
P3 | Rainfall | 0.57 | 0.284 | 0.75 | 0.141 | 0.91 | 0.028 | 0.95 | 0.012 | |
Forest | -0.66 | 0.224 | -0.92 | 0.025 | ||||||
By drainage area | A1 | Rainfall | 0.96 | 0.002 | 0.97 | 0.001 | 0.68 | 0.105 | 0.70 | 0.119 |
Forest | -0.36 | 0.489 | -0.87 | 0.434 | ||||||
A2 | Rainfall | 0.93 | 0.006 | 0.96 | 0.002 | 0.69 | 0.099 | 0.79 | 0.060 | |
Forest | -0.61 | 0.196 | - | 0.37 | 0.470 | |||||
A3 | Rainfall | 0.64 | 0.046 | 0.78 | 0.023 | 0.72 | 0.023 | 0.85 | 0.008 | |
Forest | 0.32 | 0.435 | -0.16 | 0.712 | ||||||
By rainfall and drainage area | AP1 | Rainfall | 0.95 | 0.003 | 0.96 | 0.003 | 0.54 | 0.213 | -0.70 | 0.118 |
Forest | -0.30 | 0.562 | -0.21 | 0.677 | ||||||
AP2 | Rainfall | 0.59 | 0.071 | 0.50 | 0.204 | 0.88 | 0.002 | 0.94 | 0.001 | |
Forest | 0.17 | 0.680 | -0.87 | 0.005 | ||||||
AP3 | Rainfall | 0.61 | 0.152 | 0.73 | 0.098 | 0.72 | 0.077 | 0.82 | 0.044 | |
Forest | -0.29 | 0.577 | -0.45 | 0.371 |